China Keeps Open Sourcing Its AI Models. Why?

Moonshot AI releases Kimi 2 with open-source code

It’s not intuitive but the evidence is considerable.

Conventional thinking looks at technology in China as sitting behind a great opaque wall of state surveillance and oppression. Just ask most people. Many in the US, and especially DC, will tell you that Chinese technology can’t be trusted from TikTok to Deepseek. I’m not debating the threat emanating from Chinese technology because there is significant evidence to suggest the collection of data by the Chinese Communist Party (CCP). But one of the painfully true realities in international relations is that two things can be true at the same time.

On July 11, 2025, news broke of the latest artificial intelligence (AI) model release from the Chinese company Moonshot AI. The one trillion parameter model, with 32 billion activated parameters, is called Kimi K2. According to early testing, Kimi K2 is outperforming GPT-4 along key benchmarks. It’s also open source and completely free.

Kimi K2 is the next in a line of Chinese AI algorithms that have been released open source creating an unmistakable trend. This is hard for Western analysts to make sense of because China has been historically mum with some its other tech such as its development of quantum computing and its surveillance technology. So, why AI? Why is China, that has explicitly prioritized global leadership in AI, willing to allow (or compel) its companies to release their valuable algorithms open source? There are theories, but Western analysts often fail at answering questions like these because China thinks differently. China is almost always playing a different game than we are, and this is no exception. It is also not an accident. China is buying something with these releases.

---

A Trend Worth Watching

Maybe you haven’t been watching the long trend of Chinese AI releases over the last 5 years. Maybe you’ve been busy? Maybe you are working on the next great American novel. I don’t know your life. But whatever your excuse was, it died when Deepseek R1 landed in January 2025. Its release made international news and sent panic shivers down the spines of AI investors and companies. If that was the first you heard of Chinese AI development, you aren’t alone. Moonshot AI’s Kimi K2 release is the next in what has been a long line of open-source releases from Chinese companies. Here is a list of the major companies, algorithms, and their release dates (list not intended to be exhaustive).

There are 27 algorithms from 8 companies on that list. The US list is below, and it is similarly long, but there are some striking omissions. You’ll not find models from Perplexity on the list. You’ll also not find a release from OpenAI since 2022. To its credit, Meta has been open sourcing their Llama models consistently between 2016 and 2024. But many of the “world leading” and most valuable US companies have been trending away from open-source model releases.

This clearly shows an important trend:

At least recently, Chinese companies are being more transparent with their AI development than US companies.

This must leave us asking why. Why is a country that most of us associate with developing technology behind a wall so willing to open-source technology it directly prioritizes?

---

The Curious Case of the Open-Source AI

In 2024, the MIT Technology Review published an article on this topic and it came to a couple of conclusions:

1. For Chinese cloud companies, open-source AI projects are a play to gain more market share in their cloud business.

2. Open-source AI speeds up their path to commercialization.

I don’t doubt these conclusions, but there are some deeper tactics at play.

The Trust Gap

China has been pushing back against a trust deficit that is impacting its ability to continue to globalize its technology sector. There are probably people reading this right now that are already thinking “I’ll never use one of those AI systems on that list above.” That’s fine and that’s your choice, but Chinese companies and the CCP also know that you feel that way. If China hopes to forge a path to true global AI dominance, it will have to bridge the trust gap. At least to some extent, it did this with TikTok. The threats from TikTok were made very public, yet most users are still happy to use the app. Also their choice, but it shows how the trust gap can be bridged.

The trust gap is not just about market share (but it definitely is). Gaining market share in technologies like this is also a path to both soft power and hard power. Think about the internet ecosystem from applications to services to hardware. Depending on your view of the world, you may have missed that this ecosystem is EXTREMELY US-centric. As Americans, we like it this way. It gives us all sorts of advantages from monetization of data and platforms to the safe conduct of cyber operations on US-based infrastructure. China doesn’t have this, but it knows the value. If it hopes to dominate AI the way it claims, it needs to bridge the trust gap both for economic growth and for geostrategic advantage.

Check out our AI! No, really. Check it out. Here’s the entire code base on GitHub.

Misdirection

The scarier possibility is that all this open-source AI is a misdirection. Maybe it sounds alarmist to you. Maybe it makes perfect sense to you. However you feel about it, the US is hardly in a position to discount the motives of its adversaries. Deepseek, with its ability to operate on the significantly less powerful and non-sanctioned chips, surprised everyone for the wrong reasons. We should not have been scratching our heads on how on earth China pulled ahead. We should have been looking at their innovation and on how they are changing the rules around computation capacity. Seen another way, we should have been asking ourselves why a country whose government absolutely has the ability to control the actions of private companies and who is historically fond of massive global leads in specific industries would want this to happen. Make no mistake. This was a choice. Not happenstance.

The conspiratorially minded might say that this move of “openness” is really a way to build trust in Chinese tech at a moment when they are simultaneously developing something else. The Chinese were much less transparent when they were developing their hypersonic missiles in 2021. They have not been as transparent about their quantum computing progress. So, why AI?

In long-term intelligence operations, the goal is to create a world that your subjects think is real but is controlled by you. It is hardly a stretch to imagine that China is strategically concealing its progress on other technology innovations.

Kimi K2 is an impressive piece of engineering and should be recognized for its innovations. While that is happening, we should also look at the trends. Kimi K2 is the latest in a long line of open-source models from Chinese companies at a time when US companies are less open with their models (Meta notwithstanding). We should be asking ourselves why that is. If the answer you come up with is anything other than a strategic combined geopolitical and economic play, you are probably wrong. There is a risk and potential cost of open sourcing these cutting-edge models. China wouldn’t take that risk for nothing.

Dig through the code if that’s your thing. Pay attention to its performance against benchmarks relative to ChatGPT if you’d rather. But think about why you are able to do that in the first place.

Nick Reese is the cofounder of Frontier Foundry and an adjunct professor of emerging technology at NYU. He is a veteran and a former US government policymaker on cyber and technology issues. Visit his LinkedIn here.